Epic and Microsoft partner up

AI chatbot technology could transform medical records

by

Microsoft CEO Satya Nadella did not phone it in.

He did not Zoom it in from corporate headquarters in Redmond, Washington. Nope, he was there in the flesh, at Epic Systems’ 1,700-acre campus in Verona in late August, celebrating the growing partnership between Dane County’s prize tech company and the second most valued business on Earth.

That would be Microsoft at $2.5 trillion (or so) in market valuation, which puts it behind sui generis Apple and ahead of oil-rich Saudi Aramco. If you were looking for more evidence of Epic’s growing clout in the lucrative world of medical information technology, this was it. Nadella had come to Verona.

He was one of 6,700 Epic customers, partners, vendors and observers watching the annual User Group Meeting, in addition to Epic’s 14,000 employees. “Just by showing up in Verona, Satya was telling the world this partnership is serious,” says Tom Erickson, founding director of UW-Madison’s School of Computer, Data and Information Sciences.

“Epic has not traditionally partnered with anybody in a big way,” he says. “This is new for them.” The partnership’s gains in clinician productivity and quality care, Erickson adds, are “potentially monumental.”

John Neis, a longtime Madison tech investor, is only a bit more measured as he explains that Epic and Microsoft are focused on one of the biggest challenges in healthcare: “the inordinate amount of time consumed documenting physician-patient interactions.”

“The cost savings should be meaningful, the improved productivity will help address the worker shortage, and job satisfaction will climb by simplifying mundane tasks,” Neis predicts. Then adds: “But perhaps the most intriguing aspect of the relationship is the fact that Epic and Microsoft decided to have a relationship.”

Neis points out that Epic’s history of designing software in-house is often credited as the reason why its programs integrate so well. Deciding to partner with an industry leader like Microsoft was simply recognition, Neis says, that the artificial intelligence “space is moving so fast” that Epic needed help. 

Betting that generative artificial intelligence — software that can be trained to dynamically analyze and sum up complex data with the seeming acuity of a Harvard savant — represents a reordering of life as we know it (like the paradigm shifts ignited by the onset of the internet, smartphones and GPS), Microsoft has invested at least $10 billion for a minority stake in the OpenAI startup. Those brainiacs are the creator of the ChatGPT chatbots that everyone is talking about.

Harnessing the power of AI in the generation of medical records has implications not just for doctors, patients and the medical system as a whole, but for Epic’s own breakthrough ambition: creation of the largest database of clinical information in the United States. Drawn from more than 200 Epic-affiliated hospital systems and clinics, the Cosmos database can sift anonymous “de-identified” information from more than 217 million patients in nine billion medical encounters as recorded on Epic’s ubiquitous MyChart patient portal. Still in its early stages of utilization, Cosmos is expected to be a treasure trove for clinicians and medical researchers. 

The User Group Meeting bubbled over with other good-news reports about collaborations and new programs. Epic, for example, was launching a new app store for third-party developers. There were new programs for small and large partners. And speaking of partners, Epic was very much open to collaborating with vendors who had already proven their mettle by working well with Epic customers.

For a company that in years past was sharply criticized for hogging data and resisting connection with other EHR companies and third party developers, Epic was all but announcing a new Open Door policy.

That came out loud and clear when Nadella took the stage with Epic’s senior vice president Sumit Rana for what was billed as a fireside chat at Epic’s 11,400-seat Deep Space Auditorium.

Rana described Epic’s breakthrough AI moment with palpable excitement…and a tincture of humor. It was winter break 2022. His wife was in India visiting family, and he was left to his own devices with the two kids and Baxter the Goldendoodle. It was late at night. The kids were on their own, doing what kids do. He was on his own doing what code writers do — writing code as he checked out an early iteration of ChatGPT.

And he got it! He saw how artificial intelligence could potentially reduce a doctor’s administrative workload by hours a day. It could automatically, in a matter of seconds, generate a report on the patient’s session with her doctor.

It should be noted that Microsoft was already embedded in Epic software. Microsoft’s speech-recognition subsidiary, Nuance Communications, provides Epic with real-time note-taking capability and other documentation of patient sessions. That night Rana was on the phone with Seth Hain, Epic’s senior vice president for research and development, discussing how generative AI could enhance the program.

They would jump on the project when they returned to work post-holiday, but as Rana explains later in an interview, Epic’s programmers didn’t want to wait. “They wanted in on the action right away.”

As for the reaction from Epic founder Judith Faulkner, Rana recalls she urged him onward. If he felt an urgency, the boss felt it even more. “She asked why can’t this be done now? I said we have to program it and…make it available in a safe and secure way. Stuff like that.”

“Stuff like that” means not just programming safety and security, but also subtle attitudinal pre-sets like empathy and racial sensitivity. These are huge issues for the AI world. As is a more basic question raised by medical researcher Brian Arndt: Will AI actually save a meaningful amount of time for doctors and nurses?

He sees problems ahead.

The hard fact of the matter is that the migration of medical records — richly subsidized by the federal government to the tune of $35 billion — from old-school manila folders to electronic versions has had its problems. Perhaps most notable is how EHRs swamped doctors and nurses in a tsunami of documentation and contributed significantly to the sense of burnout among many clinicians.

A 2017 study of family practice physicians led by Arndt and the University of Wisconsin School of Medicine and Public Health found that doctors over a three-year period spent more than half of their work day grappling with EHRs during and after clinic hours.

It added up to an astonishing 5.9 hours of computer busy work in an average 11.4-hour work day. Arndt, who is medical director of the UW Health Verona Clinic as well as a primary care physician for a large panel of patients, plaintively concluded: “It’s imperative to find ways to reduce documentation burden on physicians.”

Artificial intelligence, if it fulfills the sky-high expectations, could be the silver bullet that does just that: In a matter of seconds the chatbot spills out a summary of the patient encounter that the doctor accepts, modifies or rejects.

“The expectation is that a clinician will review everything for accuracy,” Nadella assured the Epic crowd. “Nothing is done automatically by the AI.”

Duly noted.

Rogue AI — as memorably portrayed in movie director Stanley Kubrick’s 1968 opus 2001: A Space Odyssey — haunts the field to this day. Malevolent coding and inexplicable “hallucinations” in AI programming are real worries. And so is the mundane time-suck of MyChart communications.

“I’m proud to be part of a health system that facilitates communications between patients and primary care physicians,” Arndt says in an email. “However, we’ve seen about a 50% increase in MyChart message volume and a 20% increase in prescription-related messages in our inbox over the past four to five years.”

He acknowledges the promise generative artificial intelligence shows for speedily documenting patient visits and responding to routine queries on MyChart. The problem is that the volume of these messages is ever mounting while the patient load remains steady.

Arndt worries that “the time and attention primary care physicians place on reviewing the AI-generated material will diminish rapidly as a simple survival skill to try to just get through the day’s workload.

“As a result, the AI-generated material’s accuracy cannot be guaranteed and will not improve unless we have sufficient time to provide the scrutiny and due diligence needed to make sure the AI tools improve over time.”

This is a somber forecast and not at all what AI enthusiasts predict.

Though it’s largely forgotten now, Epic did strike up a major partnership with the Netherlands-based Philips Medical Systems in the early aughts. “It completely blew up,” says Brendan Keeler, who worked for Epic in the Netherlands after the partnership ended. “This scarred Epic. It burned Epic so badly that they swore no more strategic partnerships.”

And it reinforced what has always been an Epic verity, per the founder: “that mergers, partnerships, acquisitions were not in our DNA,” says Keeler, who is now head of product for Flexpa, a San Francisco tech company that simplifies patient control of health records.

Keeler (who blogs at Substack as Health API Guy) draws a distinction between building customer relationships and plotting strategic partnerships. He says Epic is good about collaborating with vendors that its clinics and hospitals work well with. That’s an aspect of the attentive customer-first service that Faulkner believes central to Epic’s success.

“To watch Epic now…do some real partnerships is a big deal,” says Keeler. “It’s the first time since the Philips thing they’ve gotten over that wound.” He says Epic is looking for “best in class” partners that do tasks that Epic has chosen not to do itself.

In particular, Keeler points out the company wants to avoid any work that could enmesh Epic in a costly and lengthy federal review. This means not doing medical-device development, where human safety always triggers a cautious review, as does, it turns out, using AI in what the government classifies as “Software as a Medical Device.”

“The pre-market review can take years instead of months,” Keeler says. “Epic does not want any of that heat.” So it smartly partners with Microsoft for this work.

Rana, for his part, makes no mention of the Philips history in our interview and demurs when it’s suggested that Epic has changed its tune on partnerships. He frames it this way: The upside for artificial intelligence is so great that Epic cannot afford to ignore it. That’s why partnering with Microsoft and other “best in class” companies makes so much sense for Epic as the leading health records purveyor in the country.

“When I think about paradigm-shifting innovations in my lifetime, there’s one thing in common: There is no one company that could have done it all,” Rana told the user-group attendees. “It took many companies with many skill-sets to realize the potential. This is a similar moment. It’s going to take the village” to realize AI’s extraordinary potential.

Rana said this with the passion of a man who has seen the light. “Microsoft has been wonderful to work with,” he enthused. “It’s been a fun and intense collaboration combining advanced engineering and cutting-edge research. Together we have been pushing the realm of possibility.”

But Rana also reinforced his broader point about “the village” — that Epic would remain “vendor agnostic“ and work with other health-tech companies. Microsoft though would be a “cornerstone” partner — in the elite category reserved for Epic’s most favored allies.

Nadella, like Rana, was filled with enthusiasm for the collaboration. “It’s fantastic,” he declared. “I was keenly listening to what Judy [Faulkner] had to say about all the capabilities you build at Epic, and then how the [healthcare] community extends it and shares data.”

He called it “a virtuous cycle” that unites all the healthcare players in a common mission of improving medical care.

Nadella has a long history with generative artificial intelligence. In 2016, he wrote an influential essay for Slate in which he argued the partnership of humans and mAachines could solve society’s greatest challenges “like beating disease, ignorance and poverty.”

He championed preserving the human element in running AI. He wrote that artificial intelligence should be deployed to maximize efficiencies without destroying the dignity of people. “It should preserve cultural commitments, empowering diversity,” he wrote.

Michael Zimmer, director of Marquette University’s Center for Data, Ethics, and Society, says his impression is that Epic and Microsoft are moving slowly and deliberately with AI “I hope they are using that time to do this properly,” he says. “This is not a space where they want to ‘move fast and break things,’” as Facebook’s leaders once bragged.

“There is a whole lot of reputational harm that can happen if they get it wrong,” Zimmer adds.

Epic is already highly protective of its Cosmos data, as Caleb Cox, the company’s lead data scientist, details in an interview. The database resides on a secure federally approved channel. Only researchers associated with Epic-connected institutions are granted access. (There is no fee, by the way.) Epic explicitly bars commercial and marketing use of the de-identified patient information. Several Epic rivals, it is said, are selling their own anonymized data for commercial use.

Most surprising, Epic is not asserting ownership of the Cosmos data. Instead, it says ownership resides with the health systems that contribute the information.

As for AI in the exam room, an obvious question arises: What if a patient wants sensitive information kept out of the meeting notes? Says an Epic rep: Just ask that the ambient listening be turned off. “A doctor can pause the recording if there’s something they don’t want to be captured,” she explained.

Meanwhile, Epic and its allies are digging in to fight a proposed federal rule that aims to strengthen privacy rights. Faulkner told the users group that the rule is well-meaning, but technically problematic, and the goal can be reached by means other than hiding information. Epic’s critics and rivals, one suspects, will pounce on this issue.

The fireside chat ended with both Nadella and Rana sounding like halftime coaches. First the frank recognition that the challenge was great and that working with AI ran the risk of it being twisted to evil by cyber criminals and devious state actors.

Rana said that while artificial intelligence was going to be difficult, it would be big “and I can’t wait to see what we together accomplish.” Nadella replied that working with Epic was “fantastic, obviously,” given its core ambition to improve healthcare.

“I’m grounded in the fact that any difference we can make to health outcomes is going to help all of us across our global society,” the Microsoft leader said. “To me this is where it gets real. Technology for technology’s sake is not that interesting. But technology applied in a meaningful way on society’s biggest problems…is going to be the most impactful thing. So I am thrilled to be partners with you.”

Judging by the Epic crowd’s noisy response they felt the same about him.